@hive-mind/core

Local-first memory substrate for AI agents. The engine that powers the hive-mind project.

What's inside

• FrameStore — I/P/B frames (intra / predicted / bidirectional) with automatic compaction
• HybridSearch — FTS5 full-text + sqlite-vec vector search fused via Reciprocal Rank Fusion
• KnowledgeGraph — entities and bitemporal-valid relations
• IdentityLayer — persistent user identity
• AwarenessLayer — active-task and short-horizon state
• SessionStore — group-of-pictures conversation grouping
• Harvest pipeline — adapters for ChatGPT, Claude, Claude Code, Gemini, Perplexity, PDF, Markdown, plain text, URL, and universal formats

Install

npm install @hive-mind/core

Optional peers (only needed for specific harvest adapters and embedding providers):

npm install @huggingface/transformers pdf-parse

Quick use

import { MindDB, FrameStore, HybridSearch, KnowledgeGraph } from '@hive-mind/core';

const db = new MindDB('~/.hive-mind/my-project.mind');
const frames = new FrameStore(db);
const search = new HybridSearch(db);

frames.insert({
  frame_type: 'I',
  content: 'User prefers TypeScript over JavaScript',
  importance: 'important',
  source: 'user_stated',
});

const results = await search.query('programming language preferences');

License

Apache 2.0. See LICENSE and NOTICE.

Part of hive-mind

Full docs, architecture diagram, and the MCP server live at the monorepo root.